{"type":"rich","version":"1.0","provider_name":"Transistor","provider_url":"https://transistor.fm","author_name":"Daybreak","title":"This startup ranked AI models. They all landed in the danger zone","html":"<iframe width=\"100%\" height=\"180\" frameborder=\"no\" scrolling=\"no\" seamless src=\"https://share.transistor.fm/e/8351bf82\"></iframe>","width":"100%","height":180,"duration":738,"description":"India's best AI models are confidently wrong. Not occasionally — structurally. If you put two unrelated ideas into a prompt, the model will usually invent a connection rather than admit that none exists.In this piece, The Ken's Debanjali Biswas traces what a five-month study of leading AI models — from OpenAI, Anthropic, and Google — actually found about how they reason. The results landed almost every model in what researchers are calling the \"danger zone\", which shows high confidence and low accuracy.This is a read aloud of Debanjali's original story, by Rachel Varghese, on Daybreak.📖 Read the full story on The Ken: This startup ranked AI models. They all landed in the danger zone","thumbnail_url":"https://img.transistorcdn.com/uPitovxKRYBGX6AWg9UrET6s3nAdkS-Ci9uZvsZj7vk/rs:fill:0:0:1/w:400/h:400/q:60/mb:500000/aHR0cHM6Ly9pbWct/dXBsb2FkLXByb2R1/Y3Rpb24udHJhbnNp/c3Rvci5mbS85ODhl/ZWM4NmEwZTcxZjZk/MDRlYjAzNTNkMjJi/ZGQ2YS5wbmc.webp","thumbnail_width":300,"thumbnail_height":300}